Newman's Short Proof of the Prime Number Theorem

Author(s): D. Zagier
Source: The American Mathematical Monthly , Oct., 1997, Vol. 104, No. 8 (Oct., 1997),
pp. 705-708
Published by: Taylor & Francis, Ltd. on behalf of the Mathematical Association of
America

Stable URL: https://www.jstor.org/stable/2975232

REFERENCES
Linked references are available on JSTOR for this article:
https://www.jstor.org/stable/2975232?seq=1&cid=pdf-
reference#references_tab_contents
You may need to log in to JSTOR to access the linked references.

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms

Taylor & Francis, Ltd. and Mathematical Association of America are collaborating with JSTOR
to digitize, preserve and extend access to The American Mathematical Monthly

This content downloaded from

148.213.177.169 on Thu, 11 Feb 2021 00:21:08 UTC
All use subject to https://about.jstor.org/terms
PriRemcealNumbe
that threThenotaotreiomn .f(7r(x) g(x)x)as("fx aoo.nd g are asymptotical y equal")
Newman's Short Proof of the Prime
Number Theorem

D. Zagier
Dedicated to the Prime Number Theorem on the occasion of its 100th birthday

The prime number theorem, that the number of primes < x is asymptotic to
x/log x, was proved (independently) by Hadamard and de la Vallee Poussin in
1896. Their proof had two elements: showing that Riemann's zeta function ;(s)
has no zeros with Sc(s) = 1, and deducing the prime number theorem from this.
An ingenious short proof of the first assertion was found soon afterwards by the
same authors and by Mertens and is reproduced here, but the deduction of the
prime number theorem continued to involve difficult analysis. A proof that was
elementaty in a technical sense-it avoided the use of complex analysis-was
found in 1949 by Selberg and Erdos, but this proof is very intricate and much less
clearly motivated than the analytic one. A few years ago, however, D. J. Newman
found a very simple version of the Tauberian argument needed for an analytic
proof of the prime number theorem. We describe the resulting proof, which has a
beautifully simple structure and uses hardly anything beyond Cauchy's theorem.

means that lim<,Ot(x)/g(x) = 1, and that O(t) denotes a quantity bounded in

absolute value by a fixed multiple of f. We denote by 7T(X) the number of primes
< x.

log x

We present the argument in a series of steps. Specifically, we prove a sequence

of properties of the three functions

;(s) = E sS 4)(S) = E s , #(x) = , logp (s E ¢, x E );

n=l n p P pvx

we always use p to denote a prime. The series defining ;(s) (the Riemann zeta-
function) and 4>(s) are easily seen to be absolutely and locally uniformly conver-
gent for S(s) > 1, so they define holomorphic functions in that domain.

(I). ;(s) = Hlp(l _ p-S)-1 for S(s) > 1.

Proof: From unique factorization and the absolute convergence of ;(s) we have

;(s) 5£ (2 23r3 * ww ) = [l ( r F ) = I1 1 _ -5 ( (s) > 1) .

r2, r3, * - 20 P r20 P p

(II). ;(s)- 1 extends holomotphically to M(s) > 0.

1997] NEWMAN'S SHORT PROOF OF THE PRIME NUMBER THEOREM 705

This content downloaded from

148.213.177.169 on Thu, 11 Feb 2021 00:21:08 UTC
All use subject to https://about.jstor.org/terms
 J nS - xS J dx = sJ s uS+z dx < n<um<anX+l | uS+I | = n(5)+
Proof: For M(s) > 1 we have

1 E 1 JC° 1 E Jn+lf 1 1 )

The series on the right converges absolutely for M(s) > O because

n+l t 1 1 i n+l x du I s I Isl

by the mean value theorem.

(III). #(x)= O(x).

Proof: For n E N we have

( ° ) ( 2n ) 2 ( n ) 2 I1 p = e(2n)- o(n)

and hence, since #(x) changes by O(log x) if x changes by 0(1), #(x)-#(x/2)

< Cx for any C > log 2 and all x 2 xO = xO(C). Summing this over
x, x/2, . . ., x/2r, where x/2r 2 xO > x/2r+1, we obtain #(x) < 2Cx + 0(1).

(n). ;(S) + o and 42(s) - 1/(s - 1) is holomorphic for M(s) 2 1.

Proof: For R(s) > 1, the convergent product in (I) implies that ;(s) 7s 0 and that

_ ; ( ) = E g P = q>(s) + E g P g

The final sum converges for M(s) > 2v S° this and (II) imply that <>(s) extends
meromorphically to R(s) > -, with poles only at s = 1 and at the zeros of ;(s),
and that, if ;(s) has a zero of order ju at s = 1 + ior (or E 1R, or 7s O) and a zero of
order v at 1 + 2ior(so ,u, v 2 0 by (II)), then

lim e<)(1 + e) = 1, lim e<)(1 + e _ ior ) = -u, and lim e<)(1 + e + 2ior ) = - v.

The inequality

2 (2 + r)4)(1 + e + ira) = , gP (pia/2 +p-io.X2)4 0

then implies that 6-8 ju - 2v 2 O, so ju = O, i.e., ;(1 + ior) 7s O.

(V) | ( )2 dx is a convergent integral.

Proof: For R(s) > 1 we have

log p w d(x) w (x) w

P Ps 11 Xs sl1 X5+1 dx = sl e-st t? (et) dt.

Therefore (V) is obtained by applying the following theorem to the two functions
f(t) = (et)e-t-1 and g(z) = b(z + 1)/(z + 1)-1/z, which satisfy its hy-
potheses by (III) and (IV).

NEWMAN'S SHORT PROOF OF THE PRIME NUMBER THEOREM [October

706

This content downloaded from

148.213.177.169 on Thu, 11 Feb 2021 00:21:08 UTC
All use subject to https://about.jstor.org/terms
(VI). #(x) x.
Analytic Theorem. Let f(t) (t 2 O) be a bounded and locally integrableff>nction and
suppose that the function g(z) = t t(t)e-Zt dt (R(z) > O) extends holomorphically

to M(z) 2 O. Then J f (t) dt exists (and equals g(O)).

Proof: Assume that for some A > 1 there are arbitrarily large x with #(x) 2 Ax.
Since # is non-decreasing, we have

AX (t) - t AX)X - t A A - t
1 t2 dt 2 | t2 dt = J1 t2 dt > O

for such x, contradicting (V). Similarly, the inequality #(x) < Ax with A < 1
would imply

x#(t)-t xAx-t 1A-t

IAX t dt < | 2 dt
again a contradiction for A fixed and x big enough.
The prime number theorem follows easily from (VI), since for any e > 0

# ( x) = , log p < E log x = qr ( x) log x,

p<x p<x

#(x) 2 E logp 2 , (1 - e)logx

x 1 e < p <XX x <p <x

= (1 - e) log x[X7 (x) + o(xl e )] .

Proof of the Analytic Theorem. For T > O set gT(z) = JTt(t)e-t dt. This is

clearly holomorphic for all z. We must show that limToOgT(0) = g(O).

Let R be large and let C be the boundary of the region {z E ¢l Izl < R,
B(z) 2-b}, where 6 > 0 is small enough (depending on R) so that g(z) is
holomorphic in and on C. Then

g(°) gT(0) 2gri Jc(g(Z) gT(z))e (1 + R2 ) z

by Cauchy's theorem. On the semicircle C+= C n {R(z) > o} the integrand is

bounded by 2B/R2, where B = maxt 2ol t(t)l, because

JTZ JT S ( Z )
and

e ( 1 + R 2 )- = e ( ) * R 2
Hence the contribution to g(0)-gT(0) from the integral over C+ is bounded in
absolute value by B/R. For the integral over C_= C rA {S(z) < 0} we look at
g(z) and gT(Z) separately. Since gT is entire, the path of integration for the
integral involving gT can be replaced by the semicircle C' = {z E ¢l lZl = R,
X(z) < O}, and the integral over C' is then bounded in absolute value by 2FrB/R

1997] NEWMAN S SHORT PROOF OF THE PRIME NUMBER THEOREM 707

This content downloaded from

148.213.177.169 on Thu, 11 Feb 2021 00:21:08 UTC
All use subject to https://about.jstor.org/terms
Finally, the remaining integral over C_ tends to 0 as T oo because the integrand
by exactly the same estimate as before since

Ig (z) | = |Tf(t)e-zt dt < BJT | e-ztl dt = ( St(z) < O)

is the product of the function g(z)(1 + z2/R2)/z, which is independent of T, and

the function eZT, which goes to 0 rapidly and uniformly on compact sets as T > oo
in the half-plane M(z) < 0. Hence limsuplg(O)-gT(0)l < 2B/R. Since R is
arbitrary this proves the theorem. Tx

Historical remarks. The "Riemann" zeta function ;(s) was first introduced and
studied by Euler, and the product representation given in (I) is his. The connection
with the prime number theorem was found by Riemann, who made a deep study of
the analytic properties of ;(s). However, for our purposes the nearly trivial
analytic continuation property (II) is sufficient. The extremely ingenious proof in
(III) is in essence due to Chebyshev, who used more refined versions of such
arguments to prove that the ratio of #(x) to x (and hence also of 1T(X) to x/log x)
lies between 0.92 and 1.11 for x sufficiently large. This remained the best result
until the prime number theorem was proved in 1896 by de la Vallee Poussin and
Hadamard. Their proofs were long and intricate. (A simplified modern presenta-
tion is given on pages 41-47 of Titchmarsh's book on the Riemann zeta function
[T].) The very simple proof reproduced in (IV) of the non-vanishing of ;(s) on the
line M(s) = 1 was given in essence by Hadamard (the proof of this fact in de la
Vallee Poussin's first paper had been about 25 pages long) and then refined by de
la Vallee Poussin and by Mertens, the version given by the former being particu-
larly elegant. The Analytic Theorem and its use to prove the prime number
theorem as explained in steps (V) and (VI) above are due to D. J. Newman. Apart
from a few minor simplifications, the exposition here follows that in Newman's
original paper [N] and in the expository paper [K] by J. Korevaar.
We refer the reader to P. Bateman and H. Diamond's survey article [B] for a
beautiful historical perspective on the prime number theorem.

REFERENCES

[B] P. Bateman and H. Diamond, A hundred years of prime numbers, Amer. Math. Monthly 103
(1996), 729-741.
[K] J. Korevaar, On Newman's quick way to the prime number theorem, Math. Intelligencer 4, 3
(1982), 108-115.
[N] D. J. Newman, Simple analytic proof of the prime number theorem, Amer. Math. Monthly 87
(1980), 693-696.
[T] E. C. Titchmarsh, rhe Theory of the Riemann Zeta Function, Oxford, 1951.

Max-Planck-Institut fur Mathematik

GottJ2ied-Claren-StraJSe 26
D-53225 Bonn, Germany
zagier@mpim-bonn. mpg. de

NEWMAN'S SHORT PROOF OF THE PRIME NUMBER THEOREM [October

708

This content downloaded from

148.213.177.169 on Thu, 11 Feb 2021 00:21:08 UTC
All use subject to https://about.jstor.org/terms